2 * Copyright © 2010 - 2015 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
27 #include "common/gen_device_info.h"
28 #include "main/mtypes.h"
29 #include "main/macros.h"
38 union gl_constant_value
;
41 const struct gen_device_info
*devinfo
;
47 * Array of the ra classes for the unaligned contiguous register
53 * Mapping for register-allocated objects in *regs to the first
54 * GRF for that object.
56 uint8_t *ra_reg_to_grf
;
63 * Array of the ra classes for the unaligned contiguous register
64 * block sizes used, indexed by register size.
69 * Mapping from classes to ra_reg ranges. Each of the per-size
70 * classes corresponds to a range of ra_reg nodes. This array stores
71 * those ranges in the form of first ra_reg in each class and the
72 * total number of ra_reg elements in the last array element. This
73 * way the range of the i'th class is given by:
74 * [ class_to_ra_reg_range[i], class_to_ra_reg_range[i+1] )
76 int class_to_ra_reg_range
[17];
79 * Mapping for register-allocated objects in *regs to the first
80 * GRF for that object.
82 uint8_t *ra_reg_to_grf
;
85 * ra class for the aligned pairs we use for PLN, which doesn't
88 int aligned_pairs_class
;
91 void (*shader_debug_log
)(void *, const char *str
, ...) PRINTFLIKE(2, 3);
92 void (*shader_perf_log
)(void *, const char *str
, ...) PRINTFLIKE(2, 3);
94 bool scalar_stage
[MESA_SHADER_STAGES
];
95 struct gl_shader_compiler_options glsl_compiler_options
[MESA_SHADER_STAGES
];
98 * Apply workarounds for SIN and COS output range problems.
99 * This can negatively impact performance.
106 * Program key structures.
108 * When drawing, we look for the currently bound shaders in the program
109 * cache. This is essentially a hash table lookup, and these are the keys.
111 * Sometimes OpenGL features specified as state need to be simulated via
112 * shader code, due to a mismatch between the API and the hardware. This
113 * is often referred to as "non-orthagonal state" or "NOS". We store NOS
114 * in the program key so it's considered when searching for a program. If
115 * we haven't seen a particular combination before, we have to recompile a
116 * new specialized version.
118 * Shader compilation should not look up state in gl_context directly, but
119 * instead use the copy in the program key. This guarantees recompiles will
125 enum PACKED gen6_gather_sampler_wa
{
126 WA_SIGN
= 1, /* whether we need to sign extend */
127 WA_8BIT
= 2, /* if we have an 8bit format needing wa */
128 WA_16BIT
= 4, /* if we have a 16bit format needing wa */
132 * Sampler information needed by VS, WM, and GS program cache keys.
134 struct brw_sampler_prog_key_data
{
136 * EXT_texture_swizzle and DEPTH_TEXTURE_MODE swizzles.
138 uint16_t swizzles
[MAX_SAMPLERS
];
140 uint32_t gl_clamp_mask
[3];
143 * For RG32F, gather4's channel select is broken.
145 uint32_t gather_channel_quirk_mask
;
148 * Whether this sampler uses the compressed multisample surface layout.
150 uint32_t compressed_multisample_layout_mask
;
153 * Whether this sampler is using 16x multisampling. If so fetching from
154 * this sampler will be handled with a different instruction, ld2dms_w
160 * For Sandybridge, which shader w/a we need for gather quirks.
162 enum gen6_gather_sampler_wa gen6_gather_wa
[MAX_SAMPLERS
];
165 * Texture units that have a YUV image bound.
167 uint32_t y_u_v_image_mask
;
168 uint32_t y_uv_image_mask
;
169 uint32_t yx_xuxv_image_mask
;
173 /** The program key for Vertex Shaders. */
174 struct brw_vs_prog_key
{
175 unsigned program_string_id
;
178 * Per-attribute workaround flags
180 uint8_t gl_attrib_wa_flags
[VERT_ATTRIB_MAX
];
182 bool copy_edgeflag
:1;
184 bool clamp_vertex_color
:1;
187 * How many user clipping planes are being uploaded to the vertex shader as
190 * These are used for lowering legacy gl_ClipVertex/gl_Position clipping to
193 unsigned nr_userclip_plane_consts
:4;
196 * For pre-Gen6 hardware, a bitfield indicating which texture coordinates
197 * are going to be replaced with point coordinates (as a consequence of a
198 * call to glTexEnvi(GL_POINT_SPRITE, GL_COORD_REPLACE, GL_TRUE)). Because
199 * our SF thread requires exact matching between VS outputs and FS inputs,
200 * these texture coordinates will need to be unconditionally included in
201 * the VUE, even if they aren't written by the vertex shader.
203 uint8_t point_coord_replace
;
205 struct brw_sampler_prog_key_data tex
;
208 /** The program key for Tessellation Control Shaders. */
209 struct brw_tcs_prog_key
211 unsigned program_string_id
;
213 GLenum tes_primitive_mode
;
215 unsigned input_vertices
;
217 /** A bitfield of per-patch outputs written. */
218 uint32_t patch_outputs_written
;
220 /** A bitfield of per-vertex outputs written. */
221 uint64_t outputs_written
;
223 bool quads_workaround
;
225 struct brw_sampler_prog_key_data tex
;
228 /** The program key for Tessellation Evaluation Shaders. */
229 struct brw_tes_prog_key
231 unsigned program_string_id
;
233 /** A bitfield of per-patch inputs read. */
234 uint32_t patch_inputs_read
;
236 /** A bitfield of per-vertex inputs read. */
237 uint64_t inputs_read
;
239 struct brw_sampler_prog_key_data tex
;
242 /** The program key for Geometry Shaders. */
243 struct brw_gs_prog_key
245 unsigned program_string_id
;
247 struct brw_sampler_prog_key_data tex
;
250 /** The program key for Fragment/Pixel Shaders. */
251 struct brw_wm_prog_key
{
255 unsigned nr_color_regions
:5;
256 bool replicate_alpha
:1;
257 bool clamp_fragment_color
:1;
258 bool persample_interp
:1;
259 bool multisample_fbo
:1;
261 bool high_quality_derivatives
:1;
262 bool force_dual_color_blend
:1;
263 bool coherent_fb_fetch
:1;
265 uint16_t drawable_height
;
266 uint64_t input_slots_valid
;
267 unsigned program_string_id
;
268 GLenum alpha_test_func
; /* < For Gen4/5 MRT alpha test */
269 float alpha_test_ref
;
271 struct brw_sampler_prog_key_data tex
;
274 struct brw_cs_prog_key
{
275 uint32_t program_string_id
;
276 struct brw_sampler_prog_key_data tex
;
280 * Image metadata structure as laid out in the shader parameter
281 * buffer. Entries have to be 16B-aligned for the vec4 back-end to be
282 * able to use them. That's okay because the padding and any unused
283 * entries [most of them except when we're doing untyped surface
284 * access] will be removed by the uniform packing pass.
286 #define BRW_IMAGE_PARAM_SURFACE_IDX_OFFSET 0
287 #define BRW_IMAGE_PARAM_OFFSET_OFFSET 4
288 #define BRW_IMAGE_PARAM_SIZE_OFFSET 8
289 #define BRW_IMAGE_PARAM_STRIDE_OFFSET 12
290 #define BRW_IMAGE_PARAM_TILING_OFFSET 16
291 #define BRW_IMAGE_PARAM_SWIZZLING_OFFSET 20
292 #define BRW_IMAGE_PARAM_SIZE 24
294 struct brw_image_param
{
295 /** Surface binding table index. */
296 uint32_t surface_idx
;
298 /** Offset applied to the X and Y surface coordinates. */
301 /** Surface X, Y and Z dimensions. */
304 /** X-stride in bytes, Y-stride in pixels, horizontal slice stride in
305 * pixels, vertical slice stride in pixels.
309 /** Log2 of the tiling modulus in the X, Y and Z dimension. */
313 * Right shift to apply for bit 6 address swizzling. Two different
314 * swizzles can be specified and will be applied one after the other. The
315 * resulting address will be:
317 * addr' = addr ^ ((1 << 6) & ((addr >> swizzling[0]) ^
318 * (addr >> swizzling[1])))
320 * Use \c 0xff if any of the swizzles is not required.
322 uint32_t swizzling
[2];
325 struct brw_stage_prog_data
{
327 /** size of our binding table. */
331 * surface indices for the various groups of surfaces
333 uint32_t pull_constants_start
;
334 uint32_t texture_start
;
335 uint32_t gather_texture_start
;
339 uint32_t image_start
;
340 uint32_t shader_time_start
;
341 uint32_t plane_start
[3];
345 GLuint nr_params
; /**< number of float params/constants */
346 GLuint nr_pull_params
;
347 unsigned nr_image_params
;
349 unsigned curb_read_length
;
350 unsigned total_scratch
;
351 unsigned total_shared
;
354 * Register where the thread expects to find input data from the URB
355 * (typically uniforms, followed by vertex or fragment attributes).
357 unsigned dispatch_grf_start_reg
;
359 bool use_alt_mode
; /**< Use ALT floating point mode? Otherwise, IEEE. */
361 /* Pointers to tracked values (only valid once
362 * _mesa_load_state_parameters has been called at runtime).
364 const union gl_constant_value
**param
;
365 const union gl_constant_value
**pull_param
;
367 /** Image metadata passed to the shader as uniforms. */
368 struct brw_image_param
*image_param
;
371 /* Data about a particular attempt to compile a program. Note that
372 * there can be many of these, each in a different GL state
373 * corresponding to a different brw_wm_prog_key struct, with different
376 struct brw_wm_prog_data
{
377 struct brw_stage_prog_data base
;
379 GLuint num_varying_inputs
;
381 uint8_t reg_blocks_0
;
382 uint8_t reg_blocks_2
;
384 uint8_t dispatch_grf_start_reg_2
;
385 uint32_t prog_offset_2
;
389 * surface indices the WM-specific surfaces
391 uint32_t render_target_start
;
392 uint32_t render_target_read_start
;
396 uint8_t computed_depth_mode
;
397 bool computed_stencil
;
399 bool early_fragment_tests
;
400 bool post_depth_coverage
;
405 bool persample_dispatch
;
406 bool uses_pos_offset
;
411 bool uses_sample_mask
;
412 bool has_side_effects
;
415 bool contains_flat_varying
;
416 bool contains_noperspective_varying
;
419 * Mask of which interpolation modes are required by the fragment shader.
420 * Used in hardware setup on gen6+.
422 uint32_t barycentric_interp_modes
;
425 * Mask of which FS inputs are marked flat by the shader source. This is
426 * needed for setting up 3DSTATE_SF/SBE.
428 uint32_t flat_inputs
;
430 /* Mapping of VUE slots to interpolation modes.
431 * Used by the Gen4-5 clip/sf/wm stages.
433 unsigned char interp_mode
[65]; /* BRW_VARYING_SLOT_COUNT */
436 * Map from gl_varying_slot to the position within the FS setup data
437 * payload where the varying's attribute vertex deltas should be delivered.
438 * For varying slots that are not used by the FS, the value is -1.
440 int urb_setup
[VARYING_SLOT_MAX
];
443 struct brw_push_const_block
{
444 unsigned dwords
; /* Dword count, not reg aligned */
446 unsigned size
; /* Bytes, register aligned */
449 struct brw_cs_prog_data
{
450 struct brw_stage_prog_data base
;
452 GLuint dispatch_grf_start_reg_16
;
453 unsigned local_size
[3];
457 bool uses_num_work_groups
;
458 int thread_local_id_index
;
461 struct brw_push_const_block cross_thread
;
462 struct brw_push_const_block per_thread
;
463 struct brw_push_const_block total
;
468 * surface indices the CS-specific surfaces
470 uint32_t work_groups_start
;
476 * Enum representing the i965-specific vertex results that don't correspond
477 * exactly to any element of gl_varying_slot. The values of this enum are
478 * assigned such that they don't conflict with gl_varying_slot.
482 BRW_VARYING_SLOT_NDC
= VARYING_SLOT_MAX
,
483 BRW_VARYING_SLOT_PAD
,
485 * Technically this is not a varying but just a placeholder that
486 * compile_sf_prog() inserts into its VUE map to cause the gl_PointCoord
487 * builtin variable to be compiled correctly. see compile_sf_prog() for
490 BRW_VARYING_SLOT_PNTC
,
491 BRW_VARYING_SLOT_COUNT
495 * Data structure recording the relationship between the gl_varying_slot enum
496 * and "slots" within the vertex URB entry (VUE). A "slot" is defined as a
497 * single octaword within the VUE (128 bits).
499 * Note that each BRW register contains 256 bits (2 octawords), so when
500 * accessing the VUE in URB_NOSWIZZLE mode, each register corresponds to two
501 * consecutive VUE slots. When accessing the VUE in URB_INTERLEAVED mode (as
502 * in a vertex shader), each register corresponds to a single VUE slot, since
503 * it contains data for two separate vertices.
507 * Bitfield representing all varying slots that are (a) stored in this VUE
508 * map, and (b) actually written by the shader. Does not include any of
509 * the additional varying slots defined in brw_varying_slot.
511 GLbitfield64 slots_valid
;
514 * Is this VUE map for a separate shader pipeline?
516 * Separable programs (GL_ARB_separate_shader_objects) can be mixed and matched
517 * without the linker having a chance to dead code eliminate unused varyings.
519 * This means that we have to use a fixed slot layout, based on the output's
520 * location field, rather than assigning slots in a compact contiguous block.
525 * Map from gl_varying_slot value to VUE slot. For gl_varying_slots that are
526 * not stored in a slot (because they are not written, or because
527 * additional processing is applied before storing them in the VUE), the
530 signed char varying_to_slot
[VARYING_SLOT_TESS_MAX
];
533 * Map from VUE slot to gl_varying_slot value. For slots that do not
534 * directly correspond to a gl_varying_slot, the value comes from
537 * For slots that are not in use, the value is BRW_VARYING_SLOT_PAD.
539 signed char slot_to_varying
[VARYING_SLOT_TESS_MAX
];
542 * Total number of VUE slots in use
547 * Number of per-patch VUE slots. Only valid for tessellation control
548 * shader outputs and tessellation evaluation shader inputs.
550 int num_per_patch_slots
;
553 * Number of per-vertex VUE slots. Only valid for tessellation control
554 * shader outputs and tessellation evaluation shader inputs.
556 int num_per_vertex_slots
;
559 void brw_print_vue_map(FILE *fp
, const struct brw_vue_map
*vue_map
);
562 * Convert a VUE slot number into a byte offset within the VUE.
564 static inline GLuint
brw_vue_slot_to_offset(GLuint slot
)
570 * Convert a vertex output (brw_varying_slot) into a byte offset within the
574 GLuint
brw_varying_to_offset(const struct brw_vue_map
*vue_map
, GLuint varying
)
576 return brw_vue_slot_to_offset(vue_map
->varying_to_slot
[varying
]);
579 void brw_compute_vue_map(const struct gen_device_info
*devinfo
,
580 struct brw_vue_map
*vue_map
,
581 GLbitfield64 slots_valid
,
582 bool separate_shader
);
584 void brw_compute_tess_vue_map(struct brw_vue_map
*const vue_map
,
585 const GLbitfield64 slots_valid
,
586 const GLbitfield is_patch
);
588 /* brw_interpolation_map.c */
589 void brw_setup_vue_interpolation(struct brw_vue_map
*vue_map
,
590 struct nir_shader
*nir
,
591 struct brw_wm_prog_data
*prog_data
,
592 const struct gen_device_info
*devinfo
);
594 enum shader_dispatch_mode
{
595 DISPATCH_MODE_4X1_SINGLE
= 0,
596 DISPATCH_MODE_4X2_DUAL_INSTANCE
= 1,
597 DISPATCH_MODE_4X2_DUAL_OBJECT
= 2,
598 DISPATCH_MODE_SIMD8
= 3,
602 * @defgroup Tessellator parameter enumerations.
604 * These correspond to the hardware values in 3DSTATE_TE, and are provided
605 * as part of the tessellation evaluation shader.
609 enum brw_tess_partitioning
{
610 BRW_TESS_PARTITIONING_INTEGER
= 0,
611 BRW_TESS_PARTITIONING_ODD_FRACTIONAL
= 1,
612 BRW_TESS_PARTITIONING_EVEN_FRACTIONAL
= 2,
615 enum brw_tess_output_topology
{
616 BRW_TESS_OUTPUT_TOPOLOGY_POINT
= 0,
617 BRW_TESS_OUTPUT_TOPOLOGY_LINE
= 1,
618 BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
= 2,
619 BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW
= 3,
622 enum brw_tess_domain
{
623 BRW_TESS_DOMAIN_QUAD
= 0,
624 BRW_TESS_DOMAIN_TRI
= 1,
625 BRW_TESS_DOMAIN_ISOLINE
= 2,
629 struct brw_vue_prog_data
{
630 struct brw_stage_prog_data base
;
631 struct brw_vue_map vue_map
;
633 /** Should the hardware deliver input VUE handles for URB pull loads? */
634 bool include_vue_handles
;
636 GLuint urb_read_length
;
639 uint32_t clip_distance_mask
;
640 uint32_t cull_distance_mask
;
642 /* Used for calculating urb partitions. In the VS, this is the size of the
643 * URB entry used for both input and output to the thread. In the GS, this
644 * is the size of the URB entry used for output.
646 GLuint urb_entry_size
;
648 enum shader_dispatch_mode dispatch_mode
;
651 struct brw_vs_prog_data
{
652 struct brw_vue_prog_data base
;
654 GLbitfield64 inputs_read
;
655 GLbitfield64 double_inputs_read
;
657 unsigned nr_attributes
;
658 unsigned nr_attribute_slots
;
661 bool uses_instanceid
;
662 bool uses_basevertex
;
663 bool uses_baseinstance
;
667 struct brw_tcs_prog_data
669 struct brw_vue_prog_data base
;
671 /** Number vertices in output patch */
676 struct brw_tes_prog_data
678 struct brw_vue_prog_data base
;
680 enum brw_tess_partitioning partitioning
;
681 enum brw_tess_output_topology output_topology
;
682 enum brw_tess_domain domain
;
685 struct brw_gs_prog_data
687 struct brw_vue_prog_data base
;
689 unsigned vertices_in
;
692 * Size of an output vertex, measured in HWORDS (32 bytes).
694 unsigned output_vertex_size_hwords
;
696 unsigned output_topology
;
699 * Size of the control data (cut bits or StreamID bits), in hwords (32
700 * bytes). 0 if there is no control data.
702 unsigned control_data_header_size_hwords
;
705 * Format of the control data (either GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID
706 * if the control data is StreamID bits, or
707 * GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_CUT if the control data is cut bits).
708 * Ignored if control_data_header_size is 0.
710 unsigned control_data_format
;
712 bool include_primitive_id
;
715 * The number of vertices emitted, if constant - otherwise -1.
717 int static_vertex_count
;
722 * Gen6: Provoking vertex convention for odd-numbered triangles
728 * Gen6: Number of varyings that are output to transform feedback.
730 GLuint num_transform_feedback_bindings
:7; /* 0-BRW_MAX_SOL_BINDINGS */
733 * Gen6: Map from the index of a transform feedback binding table entry to the
734 * gl_varying_slot that should be streamed out through that binding table
737 unsigned char transform_feedback_bindings
[64 /* BRW_MAX_SOL_BINDINGS */];
740 * Gen6: Map from the index of a transform feedback binding table entry to the
741 * swizzles that should be used when streaming out data through that
742 * binding table entry.
744 unsigned char transform_feedback_swizzles
[64 /* BRW_MAX_SOL_BINDINGS */];
747 #define DEFINE_PROG_DATA_DOWNCAST(stage) \
748 static inline struct brw_##stage##_prog_data * \
749 brw_##stage##_prog_data(struct brw_stage_prog_data *prog_data) \
751 return (struct brw_##stage##_prog_data *) prog_data; \
753 DEFINE_PROG_DATA_DOWNCAST(vue
)
754 DEFINE_PROG_DATA_DOWNCAST(vs
)
755 DEFINE_PROG_DATA_DOWNCAST(tcs
)
756 DEFINE_PROG_DATA_DOWNCAST(tes
)
757 DEFINE_PROG_DATA_DOWNCAST(gs
)
758 DEFINE_PROG_DATA_DOWNCAST(wm
)
759 DEFINE_PROG_DATA_DOWNCAST(cs
)
760 DEFINE_PROG_DATA_DOWNCAST(ff_gs
)
761 DEFINE_PROG_DATA_DOWNCAST(clip
)
762 DEFINE_PROG_DATA_DOWNCAST(sf
)
763 #undef DEFINE_PROG_DATA_DOWNCAST
767 struct brw_compiler
*
768 brw_compiler_create(void *mem_ctx
, const struct gen_device_info
*devinfo
);
771 * Compile a vertex shader.
773 * Returns the final assembly and the program's size.
776 brw_compile_vs(const struct brw_compiler
*compiler
, void *log_data
,
778 const struct brw_vs_prog_key
*key
,
779 struct brw_vs_prog_data
*prog_data
,
780 const struct nir_shader
*shader
,
781 gl_clip_plane
*clip_planes
,
782 bool use_legacy_snorm_formula
,
783 int shader_time_index
,
784 unsigned *final_assembly_size
,
788 * Compile a tessellation control shader.
790 * Returns the final assembly and the program's size.
793 brw_compile_tcs(const struct brw_compiler
*compiler
,
796 const struct brw_tcs_prog_key
*key
,
797 struct brw_tcs_prog_data
*prog_data
,
798 const struct nir_shader
*nir
,
799 int shader_time_index
,
800 unsigned *final_assembly_size
,
804 * Compile a tessellation evaluation shader.
806 * Returns the final assembly and the program's size.
809 brw_compile_tes(const struct brw_compiler
*compiler
, void *log_data
,
811 const struct brw_tes_prog_key
*key
,
812 const struct brw_vue_map
*input_vue_map
,
813 struct brw_tes_prog_data
*prog_data
,
814 const struct nir_shader
*shader
,
815 struct gl_program
*prog
,
816 int shader_time_index
,
817 unsigned *final_assembly_size
,
821 * Compile a vertex shader.
823 * Returns the final assembly and the program's size.
826 brw_compile_gs(const struct brw_compiler
*compiler
, void *log_data
,
828 const struct brw_gs_prog_key
*key
,
829 struct brw_gs_prog_data
*prog_data
,
830 const struct nir_shader
*shader
,
831 struct gl_program
*prog
,
832 int shader_time_index
,
833 unsigned *final_assembly_size
,
837 * Compile a fragment shader.
839 * Returns the final assembly and the program's size.
842 brw_compile_fs(const struct brw_compiler
*compiler
, void *log_data
,
844 const struct brw_wm_prog_key
*key
,
845 struct brw_wm_prog_data
*prog_data
,
846 const struct nir_shader
*shader
,
847 struct gl_program
*prog
,
848 int shader_time_index8
,
849 int shader_time_index16
,
851 bool use_rep_send
, struct brw_vue_map
*vue_map
,
852 unsigned *final_assembly_size
,
856 * Compile a compute shader.
858 * Returns the final assembly and the program's size.
861 brw_compile_cs(const struct brw_compiler
*compiler
, void *log_data
,
863 const struct brw_cs_prog_key
*key
,
864 struct brw_cs_prog_data
*prog_data
,
865 const struct nir_shader
*shader
,
866 int shader_time_index
,
867 unsigned *final_assembly_size
,
870 static inline uint32_t
871 encode_slm_size(unsigned gen
, uint32_t bytes
)
873 uint32_t slm_size
= 0;
875 /* Shared Local Memory is specified as powers of two, and encoded in
876 * INTERFACE_DESCRIPTOR_DATA with the following representations:
878 * Size | 0 kB | 1 kB | 2 kB | 4 kB | 8 kB | 16 kB | 32 kB | 64 kB |
879 * -------------------------------------------------------------------
880 * Gen7-8 | 0 | none | none | 1 | 2 | 4 | 8 | 16 |
881 * -------------------------------------------------------------------
882 * Gen9+ | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
884 assert(bytes
<= 64 * 1024);
887 /* Shared Local Memory Size is specified as powers of two. */
888 slm_size
= util_next_power_of_two(bytes
);
891 /* Use a minimum of 1kB; turn an exponent of 10 (1024 kB) into 1. */
892 slm_size
= ffs(MAX2(slm_size
, 1024)) - 10;
894 /* Use a minimum of 4kB; convert to the pre-Gen9 representation. */
895 slm_size
= MAX2(slm_size
, 4096) / 4096;
903 * Return true if the given shader stage is dispatched contiguously by the
904 * relevant fixed function starting from channel 0 of the SIMD thread, which
905 * implies that the dispatch mask of a thread can be assumed to have the form
906 * '2^n - 1' for some n.
909 brw_stage_has_packed_dispatch(const struct gen_device_info
*devinfo
,
910 gl_shader_stage stage
,
911 const struct brw_stage_prog_data
*prog_data
)
913 /* The code below makes assumptions about the hardware's thread dispatch
914 * behavior that could be proven wrong in future generations -- Make sure
915 * to do a full test run with brw_fs_test_dispatch_packing() hooked up to
916 * the NIR front-end before changing this assertion.
918 assert(devinfo
->gen
<= 9);
921 case MESA_SHADER_FRAGMENT
: {
922 /* The PSD discards subspans coming in with no lit samples, which in the
923 * per-pixel shading case implies that each subspan will either be fully
924 * lit (due to the VMask being used to allow derivative computations),
925 * or not dispatched at all. In per-sample dispatch mode individual
926 * samples from the same subspan have a fixed relative location within
927 * the SIMD thread, so dispatch of unlit samples cannot be avoided in
928 * general and we should return false.
930 const struct brw_wm_prog_data
*wm_prog_data
=
931 (const struct brw_wm_prog_data
*)prog_data
;
932 return !wm_prog_data
->persample_dispatch
;
934 case MESA_SHADER_COMPUTE
:
935 /* Compute shaders will be spawned with either a fully enabled dispatch
936 * mask or with whatever bottom/right execution mask was given to the
937 * GPGPU walker command to be used along the workgroup edges -- In both
938 * cases the dispatch mask is required to be tightly packed for our
939 * invocation index calculations to work.
943 /* Most remaining fixed functions are limited to use a packed dispatch
944 * mask due to the hardware representation of the dispatch mask as a
945 * single counter representing the number of enabled channels.